5
votes
Accepted
Is there an argument against using the (reviewed) predictions of a model as ground truth to further train exactly this model?
Using the (unchecked) predictions of the model as training data is an approach known as "pseudo-labeling". It can help in certain situations, depending on the underlying structure of your ...
- 278
4
votes
Accepted
How to add negative samples for object detection?
The quick answer: yes you can, just add images without labels, just make sure that in the negative samples there are no cars or you will make the AI crazy (i.e. convergence & instability issues).
...
- 1,058
4
votes
Is there an argument against using the (reviewed) predictions of a model as ground truth to further train exactly this model?
The answer is: It depends.
What you describe is a strategy often used to save time and costs for labelling data. It is important that the data you have already labelled (the 20%) is representative of ...
- 205
3
votes
Accepted
What is the effect of mislabeled training data?
I think the crucial point here is what you precisely mean by mislabelled. Google's image classifier will likely do a 'pretty good' job of retrieving images with the given subject included, but how ...
3
votes
Do models train better if the labelling information is more specific (or dense)?
It depends on what is your ultimate goal. If your goal is to simply classify the object in the image, having more complex output won't help. Simpler output representation yields better result. If your ...
- 1,725
2
votes
How can computers beat humans at image recognition, if humans may incorrectly label the images?
When researchers claim "better than human accuracy", they are demonstrating that a computer can beat an individual human on a test. And that is because the ground truth labels are actually higher ...
- 26.5k
2
votes
What is the difference between "ground truth" and "ground-truth labels"?
Ground Truth
'Ground truth' is that data or information that you have that is 'true' or assumed to be true. That means that you have high or perfect knowledge of what it is. For example, in your image ...
- 1,678
2
votes
Accepted
Best practice for handling letterboxed images for non fully-convolutional deep learning networks?
Padding is indeed the easiest solution. And if no bias is used then masking the extra values during the loss computation is also not necessary, since it's enough to use zero as padding value.
You ...
- 4,763
1
vote
Using a pre-trained model to generate labels to data to then train a model on
If using BART is already giving you good results, why do you need a new model?
Not a rhetorical question. You might have good reasons for that. Training a model with less parameters optimized only on ...
- 4,763
1
vote
How to label unsupervised data for deep learning multi-classification
"is it okay to use another machine learning technology such as K-Means clustering to label the data?"
In computer vision there's an entire branch called automatic image annotation dedicated ...
- 4,763
1
vote
Should I train my network for classification on samples whose ground truth label is ambiguous?
This depends on the behaviour you want. If the ambiguous sample's ground truth is classified by a range of people, your network will get an average* based on that group. If it's only by one person, ...
- 1,316
1
vote
Accepted
How to handle an unbalanced dataset when training object detection algorithms?
One thing to try first is Focal Loss. This particular loss works well for classification or object detection where your dataset is unbalanced and contains many classes. In short, the loss suppresses ...
- 51
1
vote
Accepted
An online editor that allows data labeling format
The good folks behind Spacy have their paid product called Prodigy which is a data labeling tool. I haven't used it but it appears you can host it somewhere and then you would just have to send the ...
- 141
1
vote
Accepted
What is the difference between "ground truth" and "ground-truth labels"?
These two terms could easily refer to the same thing, depending on the context. For example, a lazy person could easily say something like this
We compute the loss/error between the prediction (of ...
- 37k
1
vote
Is intersection of labels acceptable in computer vision?
In my opinion, the second option will be more general. You can refer to some famous datasets for object detection task such as COCO or Pascal VOC, they usually accept the intersect annotations. As the ...
- 882
1
vote
Accepted
Is there a methodology for splitting up annotated orthophotos into smaller photos that retain the original bounding boxes?
You can reduce your photo size and scale the corresponding boxes to the new dimensions (416x416).
Or if you want to go with your technique, you can slice the image and then, check if the bounding box ...
- 848
1
vote
How to deal with a small amount of labeled samples?
In that particular competition, you can try using GAN to generate new data or adding noise to existing data. You can also use K-means algorithm. You can try using a smaller network and remove bias. ...
- 1,725
1
vote
How do I change the annotations of variable-size images after having resized the images to a fixed size?
Find the largest height and width amongst all the images. Let us call it H and W respectively. It is true that you cannot resize the images, but say if you have an image of height ...
- 562
1
vote
Accepted
How to label edited images after data augmentation?
Yes, you should label it the same. But more importantly you need to make sure that each perturbation of the image doesn't change some important character of the image.
Consider training an apple ...
- 1,991
1
vote
What is the effect of mislabeled training data?
I will break it down for you in very simple words. The accuracy will drop down as you label them wrong. In simpler words- accuracy is directly proportional on how perfect the data is labelled. If you ...
Only top scored, non community-wiki answers of a minimum length are eligible